The security whitepaper — written for the people who have to approve it.

Our design rule is the same one your evidence room runs on: custody is only real if you can prove it. Every control below exists to keep case data in agency custody, keep the record intact, and keep every action accountable.

1 · Deployment models and the trust boundary

CaseLead deploys three ways. The security story differs by model, so we state each boundary explicitly rather than averaging them into marketing language:

• Fully air-gapped (recommended for CJI, see §9). The complete platform — ingest, AI models, search, storage — runs on agency-owned hardware inside your facility with no internet connectivity. No third-party model APIs, no telemetry, no data egress of any kind.
• On-premises, connected. Same footprint on your network; outbound access is restricted by a fail-closed egress control: any outbound connection not on the agency-approved allowlist is blocked by default, and the platform functions fully with the allowlist empty.
• Cloud (demonstration and non-CJI use). Our hosted environment runs the public demo on de-identified sample data only. We do not host live CJI in the cloud offering.

2 · Data protection

• Encryption at rest. Case data resides on encrypted volumes (LUKS, AES-256 class). Encryption keys remain in agency custody; hardware-token-based unlock is supported so the evidence store stays sealed until an authorized token is present.
• Encryption in transit. All client–server traffic uses TLS 1.2+. In air-gapped deployments, traffic never leaves the agency network segment at all.
• Evidence integrity. Original evidence files are stored read-only and carry SHA-256 digests computed at ingest. Analysis, notes, and AI output are stored beside the record, never written into it. Redacted or shared copies are generated as new artifacts with an audit record.
• Data ownership. The agency owns the evidence, the derived index, and all work product — 100%, contractually. At contract exit you receive your data in exportable form and written confirmation of destruction of any vendor-held copies.

3 · Identity & access

• Role-based access control with least-privilege roles (administrator, supervisor, investigator, reviewer) and per-case assignment: users see only the cases they are assigned to, enforced server-side on every request.
• Multi-factor authentication at the application, with account lockout after repeated failures.
• Session controls. Configurable session timeout and concurrent-session limits aligned to CJIS Security Policy expectations for agency deployments.
• System-use notification. A CJIS-style login banner (CJIS Security Policy §5.5.4) is displayed at sign-in; the text is agency-editable to match your policy.

4 · Audit & accountability

• An immutable, append-only audit trail records authentication events, case and document access, administrative actions, and AI interactions.
• AI output is attributable: findings carry their source citations, and the record of what was asked and what was surfaced is retained — built for disclosure obligations and courtroom scrutiny, not just IT forensics.
• Audit data stays on-site, is retained per agency policy, and is exportable for supervisory or judicial review.

5 · AI-specific safeguards

These are product rules, stated in full on our Responsible AI page, that matter to a security and legal review:

• Grounded answers only. Answers are assembled from retrieved passages of the case record and carry citations to their sources. Output that cannot be traced to the record is not presented as fact.
• No AI-designated suspects, no legal conclusions. Investigators and prosecutors make determinations; the system surfaces and cites.
• Human-in-the-loop by design. Leads and findings are reviewed, verified, and dispositioned by sworn personnel; redactions require human span-by-span approval and produce an audit record.

6 · Software supply chain, updates & vulnerability handling

• Signed releases. Release and license artifacts are cryptographically signed (Ed25519) and verified before installation.
• Offline updates. For air-gapped deployments, updates arrive on signed physical media; installation takes an automatic pre-update backup and supports one-step rollback.
• Patching & CVEs. Security fixes are prioritized ahead of feature work; patch cadence and severity-based response times are committed in the support agreement, including delivery procedures for disconnected sites.
• Coordinated disclosure. We publish a vulnerability disclosure policy and security.txt, and we credit good-faith reporters.

7 · Backup & recovery

• Automated, encrypted backups with verified restore paths — recovery is exercised, not just configured.
• Backups remain on agency media in agency custody; retention follows your policy.

8 · Compliance status

We state certification status honestly — "documented" and "planned" mean exactly that. We are skeptical of vendors who imply certifications they don't hold, and we won't do it.

9 · Shared responsibility

CJIS compliance is a property of an entire deployment — hardware, network, facility, personnel screening, and agency policy — not of any single product. CaseLead is built to operate inside a CJIS-compliant environment and to make that posture the default: we bring the application-layer controls above and the documentation to map them; your agency brings the physical, personnel, and network controls it already operates. During procurement we produce a deployment-specific packet: your hardware sizing, your control mapping, and acceptance criteria you can test against.

10 · Questions your reviewers will ask — answered

• Who owns the data and the derived index? The agency. 100%. Including every embedding, extraction, and work product derived from your evidence.
• What happens at contract exit? Export of your data in usable form, written destruction confirmation for any vendor-held material, and license termination that does not hold your evidence hostage.
• Does anything leave the box in air-gapped mode? No. Egress is fail-closed; the air-gapped configuration has no outbound path to close.
• How do AI models update without internet? Signed offline media, carried in and verified on-site, with pre-update backup and rollback.
• How is AI-assisted work documented for disclosure/Brady review? Findings carry citations to the underlying record, AI interactions are logged in the audit trail, and the original evidence is provably unmodified — see Responsible AI.
• Is there an SLA? Yes — support tiers, patch cadence, and response times are defined in the support agreement, including procedures for disconnected sites.

← Back to caselead.ai